The trouble with TIA

Big Brother is watching you, and his name is John Poindexter.

To me, the most troubling story from 2002 is the U.S. Government’s reaction to the terrorism threat. I know we needed to tighten up the system. But in my opinion, the citizens of this country have had to sacrifice too many of our civil liberties in order for the country to thwart potential terrorists. I’ve written about racial profiling and face recognition at airports in this column. But what I’m about to write about makes face recognition seem benign by comparison. It even makes the FBI’s much-maligned Carnivore system seem wimpy by comparison.

I’m talking about Total Information Awareness (TIA), a system that will gather every piece of public data on every resident of the U.S. and analyze the data for terrorist patterns of behavior. It will gather data from public video feeds, credit card transactions, airline reservations, telephone calling records, etc. Any time we leave a data trail in public, it would be added to the database on us and fed into a piece of software. The software would then compare our transactions against common transaction patterns by terrorists. If there is a close enough match (however that’s defined), intelligence authorities would be notified. A prototype is already up and running; a full system could be in use in 2003. And the program is run by none other than John Poindexter, who was convicted of orchestrating the Iran-Contra affair under President Reagan. (The charges were later thrown out on a technicality.)

Why should this trouble us? Several of my readers have asked this question when I’ve written columns about national DNA and fingerprint databases, face recognition databases, etc. These readers say, “If you’ve got nothing to hide, why would you worry if the government knows everything about you?” Besides the little thing called the Constitution of the United States, my worry is not with the data gathering per se. My worry is with the software that analyzes the data. While it might get lucky and nab a terrorist or two, it will most likely identify more ordinary citizens as terrorists, leading to detentions without charge and other physical violations of innocent citizens’ civil liberties.

Pattern recognition is an inexact science. It takes years to program and train a pattern-recognition program to perform the tasks required of it. Examples from computer science abound. Take voice recognition, for example. In each case, a crude model was coded that compared a “typical” vocal item with a database full of typed items and chose the closest match. The problem with voice recognition is the infinite variety of human voices causes the crude matching system to choose the wrong item in the database more often than not. Over 20 years or so, voice recognition programs got more sophisticated and eventually enabled the user to train the pattern recognition system to correctly match spoken words 95 percent of the time (assuming you speak consistently). That seems to be the hard limit on the effectiveness of this application of pattern recognition. Handwriting recognition has topped out at about 90 percent accuracy (again, assuming consistent handwriting).

If it took that long to produce programs that still have a degree of error in them, how long will it take to write a program that takes all this data in and compares it with a database full of known terrorist activities? Probably not as long, especially since you need not have a narrow search that matches items in a one-to-one fashion. But it will take years and lots of testing. And the wider the net you cast (i.e., the more one-to-many, many-to-one, and many-to-many matches you accept as sufficient to pin the label terrorist on someone) the more false positives you will get. Plus, as terrorists change their tactics to avoid being fingered by TIA, Poindexter and co. will need to cast their nets ever wider. Finally, a degree of error in the 5 to 10 percent range can be expected, which will necessitate civil liberties violations.

The administration is talking about having a system up and running next year. I doubt it will happen; several groups have vowed to fight it in court and it will struggle there as Carnivore did. But they will at least attempt to deploy a system that by design, falsely accuses hundreds or even thousands of innocent citizens every year. If the administration succeeds, the terrorists will have succeeded in destroying not just our people and our buildings, but the very fabric of our society.

James Mathewson is editor of ComputerUser magazine and

Related posts

Leave a Comment